Machine Learning in Biomedical Images to Study Infection and Disease Phenotypes

Session

Poster Session One

Authors

Dr. Artur Yakimovich (1)

Affiliations

1. Görlitz

Keywords

deep learning, machine learning, segmentation, classification, plaque assay, virology, tissue culture, C. elegans, microscopy

Abstract text

Advances in Machine Learning (ML) and Deep Learning (DL) show unprecedented abilities for biomedical image analysis. We demonstrate in several studies how the definition of novel ML/DL tasks may aid in studying infection and disease phenotypes.

ML and DL are revolutionising our abilities to analyse biomedical images [1]. Among other host-pathogen interactions may be readily deciphered from microscopy data using convolutional neural networks (CNN) (reviewed in [2]). ML/DL algorithms may allow unambiguous scoring of virus-infected and uninfected cells in the absence of specific labelling [3]. Furthermore, accompanied by interpretability approaches, the ability of convolutional neural networks (CNNs) to learn representations, without explicit feature engineering, may allow for uncovering yet unknown phenotypes in microscopy [3]–[5]. However, this requires redefining conventional computer vision tasks beyond denoising, classification, and segmentation.

Here, we demonstrate how several such approaches may lead to novel observation from microscopy data. The first example is our recent tandem segmentation-classification algorithm aimed to understand the morphological markers of Caenorhabditis elegans motility and lifespan [6]. In this work, we combine This work proposes a combination of two deep learning models each performing a different task to summarise C. elegans microscopy data. The first network performs a classification task of  C. elegans lifespan or motility. The second network performs segmentation of the animal body parts. Using the classification results combined with saliency and segmentation both networks can pinpoint the specific body part of the worm which was found to be important for lifespan or motility. This approach allows for phenotype discovery directly from microscopy. In another work, we employ the CapsNet architecture equipped with a discriminator and generator [7]. This allowed us to differentiate between intracellular and extracellular Vaccinia virus particles through a classification task using the discriminator part of the architecture. Additionally, using the generator part we were able to visualise the differences between these particles [5]. Finally, we show how phenotype-centric packages can facilitate the data science work on virological plaque assay. This is achieved through an open-source python package we developed. We designed the package to provide the biologist with tools that make phenotypes as intuitive as data frames (manuscript in preparation).

Taken together, we show novel approaches to established algorithms in Computer Vision and Data Science. Applied to microscopy data these approaches allow for the extraction of observations from datasets large enough to not be suitable for manual analysis. We argue that this shows that reformulating conventional ML/DL tasks to answer biological questions may facilitate novel discoveries in Infection and Disease Biology.

References

[1]    Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, Art. no. 7553, May 2015, doi: 10.1038/nature14539.

[2]    A. Yakimovich, “Machine Learning and Artificial Intelligence for the Prediction of Host–Pathogen Interactions: A Viral Case,” Infection and Drug Resistance, pp. 3319–3326, 2021.

[3]    V. Andriasyan et al., “Microscopy deep learning predicts virus infections and reveals mechanics of lytic-infected cells,” Iscience, vol. 24, no. 6, p. 102543, 2021.

[4]    A. Pratapa, M. Doron, and J. C. Caicedo, “Image-based cell phenotyping with deep learning,” Current opinion in chemical biology, vol. 65, pp. 9–17, 2021.

[5]    A. Yakimovich et al., “Mimicry Embedding Facilitates Advanced Neural Network Training for Image-Based Pathogen Detection,” Msphere, vol. 5, no. 5, pp. e00836-20, 2020.

[6]    E. Galimov and A. Yakimovich, “A tandem segmentation-classification approach for the localization of morphological predictors of C. elegans lifespan and motility,” Aging (Albany NY), vol. 14, no. 4, p. 1665, 2022.

[7]    S. Sabour, N. Frosst, and G. E. Hinton, “Dynamic routing between capsules,” Advances in neural information processing systems, vol. 30, 2017.